WO2020220563A1

WO2020220563A1 - Auditory brainstem response automatic test method based on adaptive averaging method

Info

Publication number: WO2020220563A1
Application number: PCT/CN2019/106153
Authority: WO
Inventors: 华云峰; 吴皓; 王皓煜; 李孛; 丁旭; 黄治物; 汪雪玲
Original assignee: 上海交通大学医学院附属第九人民医院
Priority date: 2019-04-30
Filing date: 2019-09-17
Publication date: 2020-11-05
Also published as: CN110074782A; CN110074782B

Abstract

An auditory brainstem response automatic test method based on an adaptive averaging method. With regard to auditory brainstem response single records collected in batches under various test sound intensities from high to low, on the basis of calculation performed using an adaptive averaging method, the average number of records is increased gradually by means of iteration, and a signal-to-noise ratio is increased until an auditory brainstem response signal detection condition is met, the condition involving obtaining two groups of average curves by means of auditory brainstem response single records collected in groups and in batches, calculating a time lag where the maximum value of a cross-correlation function of the two groups of average curves is located, and determining, according to whether the deviation of the time lag is within a specified range, whether there is an auditory brainstem response signal with a time locking characteristic. A hearing threshold value is obtained by means of the minimum sound intensity required for detecting a signal; or an accurate sound intensity corresponding to the hearing threshold value is obtained by performing function fitting on the number of iterations used in each test sound intensity and by means of an interpolation method. The method has a high efficiency in detecting a threshold value, has an accuracy close to that of manual determination, is more objective, has a better repeatability, and can effectively reduce the number of times that an auditory brainstem response record is repeatedly collected.

Description

Automatic test method of auditory brainstem response based on adaptive average method

Technical field

The invention relates to data processing technology, in particular to an automatic test method for auditory brainstem response based on an adaptive average method.

Background technique

Auditory Brainstem Response (ABR) is a brain electrical change induced by sound stimulation. Its characteristic waveform occurs within 10 milliseconds after the stimulation and can be recorded by electrodes placed on the head of a human or animal. The non-invasive feature makes ABR widely used as a means of hearing detection in clinical practice, especially for infants and young children, people with intellectual disabilities, and intraoperative patients, who cannot complete hearing assessment through verbal communication or physical movements. In clinical practice, hearing threshold (referring to the minimum sound intensity used to induce a subject’s ABR characteristic waveform) has become one of the important indicators for evaluating hearing function.

However, due to the large changes in the signal-to-noise ratio recorded by the ABR and the large differences in waveforms at various sound intensities, the current judgment is basically based on the records after hundreds of averaging by professionally trained physicians. This subjective method of judging the threshold causes the accuracy of the results to depend on personal experience and skills, and increases the degree of dependence on professional manpower, which cannot meet the increasing clinical needs. This has caused the development of infant hearing screening nationwide. Great difficulty. This also makes it difficult to realize the automated analysis of ABR test data. In the past 30 to 40 years, people have made a series of efforts in the automation of ABR hearing test and put forward various solutions, but there is no mature and reliable one. Commercial products come out.

Disclosure of invention

The purpose of the present invention is to provide an efficient and reliable automatic test method for auditory brainstem response based on the adaptive average method.

The technical solution of the present invention is to provide an automatic test method for auditory brainstem response based on the adaptive average method. For the ABR records collected in batches under each test sound intensity, based on the calculation of the adaptive average method, the records are gradually increased through iteration Average times until the conditions for ABR signal detection are met;

The conditions for detecting the ABR signal include: after grouping the ABR records corresponding to the current sound intensity at the current iteration, calculating the average curve of each group according to the current average times, and obtaining the value corresponding to the maximum value of the cross-correlation function between the groups Time lag, according to whether the deviation of the time lag is within the specified range, it is judged whether there is an ABR signal with time-locking characteristics.

Optionally, the operation of the adaptive average method includes the following processes:

S1', the ABR record is a time curve whose input type is a repeated single record, which is randomly divided into two groups and the average curve is calculated respectively;

S2'. Calculate the cross-correlation function after the average of the two groups;

S3', obtaining the time lag corresponding to the maximum value of the cross-correlation function;

S4', compare the absolute value of the time lag with the value of k, and judge whether the time lag is within k data points:

If the absolute value of the time lag is less than the value of k, it means that the time lag deviation is within k data points, which means that the stable time lock signal is detected, and the acquisition of data is stopped, which corresponds to the detection of the ABR signal;

If the absolute value of the time lag is not less than the value of k, it means that the time lag deviation is not within k data points, indicating that the stable time lock signal is not detected, then step S5' is further executed:

S5', judge whether the maximum number of iterations under the current sound intensity is reached:

If the maximum number of iterations is reached, the acquisition of the collected data is stopped, which corresponds to the situation where the ABR signal is not detected; if the maximum number of iterations is not reached, the newly collected ABR records are further added to the currently collected ABR records, and the current sound intensity Repeat the calculation process of S1'～S5'.

The ABR record is a time curve whose input type is a repeated single record, which is judged Q times in parallel, and steps S1 to S4 are executed each time;

S1. The currently collected ABR records are randomly divided into two groups and the average curves are calculated respectively;

S2. Calculate the cross-correlation function after the average of the two groups;

S3. Obtain the time lag corresponding to the maximum value of the cross-correlation function;

S4. Compare the absolute value of the time lag with the k value to determine whether the time lag is within k data points:

If the absolute value of the time lag is less than the value of k, it means that the time lag deviation is within k data points, indicating that a stable time-lock signal is detected; if the absolute value of the time lag is not less than the value of k, it means that the time lag deviation is not in k data Within the point, it means that no stable lock signal is detected;

S5. Judge whether the stable lock signal is detected every time in Q parallel judgments:

If the stable lock signal is detected every time in Q times, stop acquiring the collected data, which corresponds to the situation where the ABR signal is detected;

If the stable lock time signal is not detected every time in Q times, step S6 is further executed:

S6. Determine whether the maximum number of iterations under the current sound intensity is reached:

If the maximum number of iterations is reached, stop acquiring the collected data, which corresponds to the situation that no ABR signal is detected;

If the maximum number of iterations is not reached, the newly collected ABR record is further added to the currently collected ABR record, and the calculation process from S1 to S6 is repeated under the current sound intensity.

S1 ", the ABR is that the input type is the average recording time of multiple curves; mean curve avgA this newly added _plus iteration, the mean curve avgA _old binding the previous iteration, the weight is calculated in accordance with the current average weights obtained _new curve avgA

Among them, avgA _new , avgA _old , avgA _plus have Q groups respectively;

M is the current number of iterations;

S2", _{new to the} current average curve avgA of group Q, calculate the cross-correlation function between each two groups;

S3", obtain the Q group time lag corresponding to the maximum value of the cross-correlation function;

S4". In the Q parallel judgment, the absolute value of each group of time lag is compared with the value of k to determine whether the time lag is within k data points: if the absolute value of the time lag is less than the value of k, it means the time lag If the deviation is within k data points, it means that the stable time-lock signal is detected; if the absolute value of the time lag is not less than the k value, it means that the time-lag deviation is not within k data points, which means that the stable time-lock signal is not detected;

S5", judge whether the stable lock signal is detected every time in Q parallel judgments:

If the stable lock signal is not detected every time in Q times, step S6" is further executed:

S6", judge whether the maximum number of iterations under the current sound intensity is reached:

If the maximum number of iterations is reached, the acquisition of collected data is stopped, which corresponds to the situation where the ABR signal is not detected; if the maximum number of iterations is not reached, the newly collected ABR test average data is further added to the currently collected ABR test average data. Repeat the calculation process of S1"～S6" under the current sound intensity.

Optionally, k is a preset fixed value, or a value obtained by calculating all data points according to a preset ratio.

Optionally, judging whether the maximum number of iterations under the sound intensity is reached includes judging whether M=M _max ; where M _max is the preset maximum number of iterations; M is the current number of iterations, and each time step S1 M is incremented by 1: if M< _Mmax , the iteration continues.

Optionally, the average number of times used when calculating the average curve of the ABR record at each sound intensity is the product of the current iteration number and the value of N; the value of N corresponds to the number of groups of ABR records newly added in each iteration.

Optionally, the hearing threshold is obtained by detecting the minimum sound intensity required by the ABR signal; or, for the number of iterations used at each sound intensity, the sound intensity corresponding to the hearing threshold is obtained by function fitting and interpolation.

Optionally, after normalizing the number of iterations used at each sound intensity, the sound intensity corresponding to the hearing threshold is obtained through Sigmoid function fitting and interpolation.

Optionally, the ABR records at each sound intensity are animal experiment data or clinical data, data obtained through real-time testing or offline storage;

The ABR record undergoes one or more of the following pre-processing: signal amplification; band-pass filtering; adjusting the time interval for collecting the ABR waveform, selecting the ABR time curve of the corresponding time interval after the sound stimulation starts as the analysis object; ABR time curve corresponding to background noise; smooth spline fitting function to remove low-frequency background noise.

In the traditional ABR test, the test sound intensity is from high to low, and each sound intensity is repeated a fixed number of times to obtain an average signal (Figure 1A). The lowest sound intensity but still with auditory brainstem response is judged by the physician as the hearing threshold ( It is believed that the brain responds above this threshold and can be heard; Figure 2A). In the previous 30-40 years, in order to reduce the misdiagnosis caused by artificial subjective judgment, the industry has been developing automatic hearing test methods, but they have not succeeded. The reason is that the quantitative analysis of the signal will be affected by the experimental conditions (such as the degree of anesthesia, electrode Placement position, etc.). Therefore, it is difficult to have an absolute value (signal-to-noise ratio, correlation coefficient, etc.) that can be considered to have a signal when it is higher than this value, and there is no signal when it is lower than this value, which can correspond to the hearing threshold (Figure 2C, x = 0 is the true threshold, but corresponds to a different y value).

Compared with the prior art, the self-adaptive average algorithm provided by the present invention is used to analyze the data generated in the hearing test based on the auditory brainstem response (ABR), which can replace the artificial and automatic acquisition of the subject’s hearing threshold (hearing threshold). The main indicator of the test), which can dynamically adjust the number of times each sound intensity is repeated. In other words, since there is no exact value that can correspond to the hearing threshold, first set a standard, and then solve the problem of how many times it needs to reach this standard on average.

For this reason, the adaptive average algorithm of the present invention gradually increases the average number of records through iterations for the ABR records collected in batches at various test sound intensities from high to low, so as to increase the signal-to-noise ratio until the conditions for ABR signal detection are met. According to the ABR signal detection conditions, two sets of average curves are obtained by randomly grouping and batch-collecting ABR records, calculating the time lag where the maximum value of the cross-correlation function is located, and judging whether there is a time lock according to whether the deviation of the time lag is within the specified range Characteristic ABR signal. The acquisition of the hearing threshold can be: the minimum sound intensity required to detect the signal, or the number of iterations used at each sound intensity, the accurate sound intensity corresponding to the hearing threshold can be obtained through function fitting and interpolation.

Use the cross-correlation function to judge, when there is a time-locked signal (ABR), the theoretical time lag is 0 (the actual calculation will deviate by k data points, the example k=1, corresponding to the solid point in Figure 3D); if there is no signal, The time lag is any value (corresponding to the hollow point in Figure 3D).

When using the adaptive averaging algorithm of the present invention, if the signal-to-noise ratio is high, the signal can be detected without averaging many times; if the signal-to-noise ratio is low, the signal can be detected only by averaging more times. The algorithm can dynamically adjust the average times. If there is no signal, no signal will be detected after averaging multiple times (we set the upper limit), as shown in Figure 3E.

On the other hand, the number of iterations required to detect the ABR signal under each test sound intensity acquired by the present invention can obtain more accurate hearing thresholds through function fitting and interpolation, improve the accuracy of threshold judgment, and effectively reduce ABR records The number of repeated acquisitions. If the test sound intensity decreases by 5dB, the traditional method only has 5dB accuracy, but the present invention can reach 1dB through function fitting (Figure 3F).

The present invention can also target two different input data types (Figure 3A and Figure 4), which are the data recorded in a single repetition and the averaged data. These two data formats cover all clinical models. All invented algorithms can be applied.

In the method of the present invention, the accuracy of the error within ±5dB is basically 100%. At the same time, since each sound intensity detection signal executes the next sound intensity, it can save up to 70% of repeated recording (Figure 5A and Figure 5B) .

In summary, the present invention proposes a calculation model of the adaptive averaging method, thereby realizing an efficient and reliable automatic test method for auditory brainstem response. The accuracy of the hearing threshold obtained by this method is close to the artificial judgment of a professional. It is more objective and repeatable. More importantly, the method of the present invention does not need to adjust the model according to the data quality during use, so it has broad application prospects in various clinical and scientific ABR tests. At the same time, it is worth mentioning that the method of the present invention can terminate the iteration according to the judgment result and feed it back to the hardware in real time, which can not only shorten the test time and avoid the waste of storage space, but also can be used as a core module for unmanned automatic ABR hearing detection. The device is expected to free clinicians completely from the task of hearing testing.

Brief description of the drawings

Figure 1A is a smoothed mouse ABR time curve and averaged data under four sound intensities;

Figure 1B shows the distribution of correlation coefficients between any two curves under four sound intensities and Gaussian distribution fitting;

Figure 2A is the ABR time curve corresponding to normal mice and mice at high risk of deafness after an average of 350 times under each sound intensity (SPL);

Figure 2B is the median of the correlation coefficients between any two records under each sound intensity;

Figure 2C is the median of the correlation coefficients near the threshold, and the correlation coefficient corresponding to the hearing threshold is approximately 0.01;

Figure 3A is a flowchart of the adaptive average algorithm when the input data type is a single record;

Figure 3B is the mouse ABR time curve after 500 averaging at various sound intensities;

Figure 3C is the result of the same set of data after the adaptive average method. The two curves corresponding to each sound intensity are the average curves used to calculate the cross-correlation function;

Figure 3D is the time lag corresponding to the maximum correlation coefficient obtained in the iteration;

Figure 3E is the number of iterations required to detect a signal under each sound intensity;

Fig. 3F is the corresponding threshold obtained by fitting the normalized iteration number through the Sigmoid function;

Figure 4 is a flowchart of the adaptive average algorithm when the input data type is averaged data;

FIG. 5A is a comparison of the deviations of threshold values obtained in these three ways when the doctor uses the traditional average method to read the test data, the doctor reads the data required by the adaptive algorithm, and the machine applies the adaptive algorithm of the present invention;

FIG. 5B is a comparison of the normalized total number of records obtained in the two methods when the physician uses the traditional averaging method to read the test data and the machine applies the adaptive algorithm of the present invention.

The best way to implement the invention

For a given sound stimulus (short tones, short pure tones are often used), the auditory system produces a series of potential responses. Auditory brainstem response (ABR) test, by recording the waveform changes of these potentials, and using various digital signal processing algorithms to extract them from various strong noise backgrounds to obtain auditory brainstem evoked potentials, which are used to evaluate the auditory conduction system An important indicator of integrity and monitoring of nervous system function.

The present invention proposes an algorithm model based on the adaptive average method, which can automatically process the data signals obtained in the ABR test, thereby realizing an automatic test method for auditory brainstem response.

First, briefly describe the traditional testing methods and the principles involved in the present invention:

In the traditional way, the doctor analyzes the original records obtained by the ABR test. For example, Figure 1A shows four sound intensities (80dB, 60dB, 40dB, 20dB), and each sound intensity is repeated a fixed number of times to test the mice. Obtain multiple time curves corresponding to the smoothed mouse ABR test (indicated in gray). The figure also shows the curves a1 to a4 corresponding to the averaged data under each sound intensity (the heterochromatic curve in gray) . Originally, based on the above data, the doctor judged which is the lowest sound intensity but still has auditory brainstem response, and used it as the hearing threshold. When the threshold is higher than the threshold, the brain responds, indicating that it can be heard.

Figure 1B shows the distribution of correlation coefficients between any two curves at each sound intensity (in this figure, it is represented as a black curve that seems to be jagged at each sound intensity), and the corresponding Gaussian distribution fitting (represented as each sound Stronger and smoother bright color curves b1 to b4).

Through the data analysis of Figure 1A and Figure 1B, it is found that when the stimulus sound intensity is less than the hearing threshold, the correlation coefficient between ABR records is symmetrically distributed around 0 (not relevant), for example, see curves b1 and b2; while the stimulus sound intensity is higher than At the hearing threshold, the correlation coefficient distribution moves in the +1 direction (positive correlation), for example, see curves b3 and b4.

Then verify the above model: using this model, the minimum correlation coefficient required to detect a stable ABR waveform can be determined through experience (see Figure 2C, which represents the median of the correlation coefficient near the threshold), which can become the threshold value Judgment criteria (described in detail below).

However, in the actual operation process, the difference in the signal-to-noise ratio of the data due to different experimental conditions (such as electrode placement, animal anesthesia, etc.) affects the modeling of the correlation coefficient, so each record needs to be calibrated to ensure it The accuracy of threshold judgment affects the practical value. In view of the above problems, the present invention proposes the use of the classic judgment criterion of the time-locked signal (that is, the time lag at the maximum cross-correlation function is 0), and adjusts the signal-to-noise ratio by changing the number of averaging, and finally realizes the dynamic detection of ABR signals. This is the adaptive average method. That is, the present invention dynamically adjusts the number of repetitions of each sound intensity, establishes a standard corresponding to the hearing threshold, and uses the adaptive averaging method to determine how many times the standard needs to be averaged to achieve the hearing threshold.

Therefore, in the specific algorithm of the adaptive average method, the present invention uses the cross-correlation function to determine whether there is a time-locked signal (ABR). When there is a signal, the theoretical time lag is 0 (the actual calculation will deviate by k data points. k=1), or when there is no signal, the time lag is any value.

In the example in Figure 3A, the data recorded by a single sound stimulus is randomly grouped and averaged (while the other example in Figure 4 is suitable for the case where only the average data can be derived, detailed below), using "the maximum value of cross-correlation function Whether the time lag is within a data record point" (due to system error, generally 1 data point <50 microseconds), to determine whether there is a time-locked ABR signal; if no signal is detected, the data volume is gradually increased through iteration until The signal appears or reaches the preset maximum value.

The actual test results of the above detection methods prove that when the stimulus sound intensity is higher than the hearing threshold, the average number of times required to meet the time lag condition is maintained at a low level; and the average number of times required to meet the threshold increases exponentially; and When it reaches the preset maximum number below the threshold, the hearing threshold is the highest sound intensity without a detected signal. It can be seen that when the adaptive averaging method of the present invention is used, if the signal-to-noise ratio is high, the signal can be detected without averaging many times; if the signal-to-noise ratio is low, the signal can be detected by averaging more times; The upper limit, if there is no signal, no signal will be detected even after averaging multiple times.

The following examples illustrate the experimental process of the method of the present invention. The steps involved are:

(1) Data acquisition and preprocessing

Animal experiment: Through the TDT RZ6/BioSigRZ system, collect mouse ears from 90 to 0dB (5dB interval) short pure tone (frequency 16 kHz, duration 3 milliseconds) excited ABR data (the original signal is amplified 20000 times, and passed 50-5000 Hz bandpass filter). Each sound intensity was recorded 500 times, the stimulation signal rate was 21 times/second, the data acquisition frequency was 21kHz, and the acquisition time interval was 0-15 milliseconds after sound stimulation.

In the TDT RZ6/BioSigRZ system, the TDT auditory evoked potential workstation is constituted by the RZ6 processor, preamplifier and BioSigRZ experimental design analysis software. Working with low impedance electrodes, needle electrodes, and surface electrodes, the workstation can record various bioelectric signals including brainstem evoked potentials and otoacoustic emissions.

Clinical: Data comes from shared records funded by NIH (www.physionet.org). The experimental condition is that ABR records 100dB to 30dB (5dB as spacing) stimulus sound (1kHz or 4kHz short pure tone, stimulation rate 24 times/sec) through a single ear. The subject's electrode placement is as follows: the recording electrode is placed on the forehead and the reference electrode is placed After stimulating the mastoid on the same side of the ear, the earth pole is placed on the opposite side of the stimulating ear. Each sound intensity is recorded 1000 times, the ABR signal is filtered by 30Hz to 3000Hz bandpass, the amplifier is amplified by 50000 times and then recorded, the sampling rate of the recording equipment is 48kHz, and the collection time interval is 0-15 milliseconds after the sound stimulation. The above-mentioned acquisition of animal experiment and clinical ABR data, as well as the recorded parameters during preprocessing are only examples, and can actually be adjusted according to specific application conditions.

(2) According to the time interval corresponding to the ABR waveform, the ABR time curve between 0-6 milliseconds (mice) or 5-15 milliseconds (clinical) after the start of the sound stimulation is selected as the analysis object. The time interval involved in this step is an empirical value and can be adjusted according to actual conditions.

(3) Background noise caused by muscle activity or breathing can be eliminated by removing time curves greater than 11 microvolts or less than -11 microvolts (this step is optional, and the parameter values involved depend on the settings of the hardware used. ).

(4) Remove low-frequency background noise by smoothing spline (smoothing spline) fitting function (this step is optional, the smoothing parameter in the example is 0.5).

(5) Take 350 curves for each sound intensity, and obtain the correlation coefficient distribution between any two sets of curves, see Figure 1A and Figure 1B. Calculated as follows:

A, B∈M, and A≠B

Among them, A, B are the data recorded in a single time; N is the number of data points recorded; μA, μB are the average curves of a single record; σA, σB are the standard deviations of a single record; M is the number of iterations.

(6) Take the median of the correlation coefficient of each sound intensity and plot, it can be seen that the correlation coefficient corresponding to the sound intensity higher than the hearing threshold is larger than the correlation coefficient below the hearing threshold. Refer to Figure 2A, which shows the mouse ABR time curve after an average of 350 times under each sound intensity (SPL) (each sound intensity corresponds to two curves: the black solid line corresponds to normal mice, such as c91, c61, c31 Etc., the gray dashed line corresponds to mice with high risk of deafness, such as c92, c62, c32, etc.); see Figure 2B, based on the data in Figure 2A, further give the middle of the correlation coefficient between any two records at each sound intensity Number of digits, the hollow points correspond to normal mice, and the solid points correspond to mice at high risk of deafness.

(7) When the clinician judges the threshold in the traditional way, the curve is aligned (see Figure 2C, which is the median of the correlation coefficients near the threshold. The gray curves d1, d3, d5, d7, d8 correspond to deafness For high-risk mice, the black curves d2, d4, and d6 correspond to normal mice), it can be seen that the correlation coefficient corresponding to the hearing threshold is approximately 0.01. However, the signal-to-noise ratios obtained by different experiments are different (correlation coefficients measured by reducing background noise), resulting in poor overlap between the obtained curves, and the accuracy of the judgment threshold will be limited.

(8) The adaptive average rule of the present invention gradually increases the number of records and uses random group averaging to calculate whether the absolute value of the time lag corresponding to the maximum value of the cross-correlation is less than k data points to determine whether there is a stable lock time Signal up to the maximum number of iterations. k is a fixed value corresponding to a fixed number of data recording points, or k is a value calculated according to a certain ratio (for example, 1% of all data points); the following uses k=1 as an example for explanation and experimental verification). Experimental data confirms that when it is greater than the hearing threshold, the absolute value of lag is maintained at a level less than k=1 (Figure 3D).

(9) The adaptive averaging method can terminate the iteration because the signal is detected or there is no signal after reaching the maximum number of iterations. If the number of iterations at each sound intensity is used as the output (Figure 3E), the Sigmoid function fits after normalization (Figure 3F),

The coefficients a and T can be obtained. According to the data verification of 1dB interval (see the inset in Figure 3F), the x value when f(x) = 0.9 is the threshold, and the result is consistent with the threshold manually judged.

As shown in FIG. 3A, in an embodiment where the input data type is a single recording, the method for automatically testing auditory brainstem responses based on the adaptive average method of the present invention includes the following processes:

S1. Divide the currently collected ABR data into two groups at random, and calculate the average curve for each group; the ABR time curve corresponding to the average curve of each group of data can be obtained (the two curves under each sound intensity in Figure 3C) ；

S4. Compare the absolute value of the time lag with k=1: if the absolute value of the time lag is less than 1, it means that a stable time-lock signal is detected (corresponding to the solid points in Figure 3D); if the absolute value of the time lag is not If it is less than 1, it means that no stable lock signal has been detected (corresponding to the hollow points in Figure 3D).

The present invention can judge multiple times in parallel, and execute the above-mentioned steps S1-S4 each time, so as to avoid the problem of "the background noise meets the required time lag under extremely accidental circumstances". In this example, three parallel judgments are made, and the three times are randomly grouped and averaged into avgA, avgB; avgA', avgB'; avgA”, avgB”, and calculate the mutual values of avgA and avgB, avgA' and avgB', avgA” and avgB” Calculate the corresponding time lag after the relationship number, and record the judgment result if the absolute value of the time lag is less than 1 (data point) as "1", and record the judgment result if the absolute value of the time lag is not less than 1 (data point) as "0" "; It further includes the following process:

S5. If the three judgments all indicate that the stable lock signal is detected under the current sound intensity, that is, the judgment results corresponding to the three judgments A, B, and C are all "1", then the data collection can be stopped, indicating that the detection is possible under the current sound intensity When the signal reaches the stability lock, the test is passed under the current sound intensity;

Otherwise, if the judgment results corresponding to three judgments A, B, and C are not all "1", then perform further:

S6. Determine whether the maximum number of iterations under the sound intensity is reached:

That is, it is judged whether M=M _max ; where M _max is the preset maximum number of iterations; M is the current number of iterations, each time step S1 is executed, M is incremented by 1: If M<M _max , the iteration continues;

If the maximum number of iterations is reached, you can stop collecting data, indicating that the stable time lock signal cannot be detected under the current sound intensity, and the test has not passed under the current sound intensity; if the maximum number of iterations is not reached, the newly collected data will be added to the original data. For N sets of data (N=50 in this example), the process of steps S1-S6 is repeated under the current sound intensity. For other sound intensities, repeat the process of steps S1-S6.

Figure 3B is the mouse ABR time curve after 500 averaging at various sound intensities; Figure 3C is the result obtained by adaptive averaging using the same set of data in Figure 3B. The black/gray lines at each sound intensity are two groups The average curve used to calculate the cross-correlation function. The average number of each sound intensity is the product of the current iteration number M and N shown in Figure 4E (in this case ×50); Figure 3D is the maximum value obtained in the iteration The time lag (absolute value) corresponding to the correlation coefficient. The solid point indicates that the stable lock time signal is detected, and the hollow point indicates that the stable lock time signal is not detected; according to Figure 3D, the solid point jumps to the hollow point corresponding to the sound intensity interval , It is known that the hearing threshold of this example is between 25-30dB; Figure 3E shows the number of iterations required to detect the signal at each sound intensity. If there is no signal for 2 consecutive times (the maximum number of iterations is reached), it means that the hearing threshold is lowered and the test is automatically terminated. The solid point in the figure is the actual test sound intensity, and the hollow point is the sound intensity that does not need to be tested (because there is no test for 2 times). After the signal is detected, the test can be terminated in principle).

As shown in Figure 5A (the left group and the right group, corresponding to animal experiment data and clinical experiment data, respectively), the doctor reads the average curve of all data in the traditional way for threshold judgment, and the doctor reads the adaptive algorithm of the present invention The data obtained by averaging the number of times is subjected to threshold judgment, and the machine uses the data obtained from the averaging number of the adaptive algorithm of the present invention to perform the threshold judgment, and the deviations of the threshold value obtained under these three methods are compared. As shown in Figure 5B (the left group and the right group, corresponding to animal experiment data and clinical experiment data, respectively), the normalized number of data (that is, the total number of records) of all the data required when the physician uses the traditional method to judge the threshold ), which is compared with the normalized number of data pieces required for threshold judgment through the adaptive algorithm of the present invention.

With reference to Figures 5A and 5B, compared with the average curve of hearing threshold judged by multiple doctors, the accuracy of the method of the present invention within ±5dB is basically 100%. In addition, compared with using the maximum number of averages for each sound intensity, the present invention executes the next sound intensity test as soon as the signal is detected for each sound intensity, thereby reducing up to 69% of the records that do not contribute to the threshold judgment. Save up to 69% of duplicate records.

The present invention can further improve the accuracy of hearing threshold judgment through function fitting and interpolation. Figure 3F shows the normalized number of iterations. The sound intensity corresponding to the hearing threshold is obtained through Sigmoid function fitting and interpolation. (Inset) ABR detection results with 10dB up and down near the threshold and 1dB interval. The threshold obtained in this example is 26dB . That is to say, if the test sound intensity decreases by 5dB, the traditional method only has 5dB accuracy, but the present invention can reach 1dB through function fitting.

For the case where only average data can be derived, in another embodiment shown in FIG. 4, the ABR automatic test method based on the adaptive average method of the present invention (input data type is averaged data) is the same as that of the previous embodiment. The difference is:

Three groups of average data collected continuously are added for each iteration (for example, each group averages 50 times; the number of groups corresponds to the number of parallel judgments), combined with the previous data, recalculates the average curve according to the weight;

For example, for the fourth iteration:

avgA (new) is the current average curve;

avgA (old) is the average curve of the previous iteration;

avgA (plus) is the newly added average curve;

For the current three sets of updated average curves, calculate the cross-correlation function between each two groups to obtain the three sets of time lags corresponding to the maximum value of the cross-correlation function, and perform the judgment of the absolute value of the time lag and k=1 (data point) . If the three sets of judgments all indicate that the stable lock signal is detected under the current sound intensity, you can stop collecting data, which means that the stable lock signal can be detected under the current sound intensity, and the test is passed under the current sound intensity; otherwise, it is further judged whether it has reached this value. Maximum number of iterations under sound intensity: If the maximum number of iterations is reached, the data collection will stop, which means that the stable lock signal cannot be detected under the current sound intensity, and the test has not passed under the current sound intensity; if the maximum number of iterations is not reached, the original Add new average data to some average data and repeat the above process. Repeat the above process for other sound intensities.

Although the content of the present invention has been described in detail through the above preferred embodiments, it should be recognized that the above description should not be considered as limiting the present invention. After those skilled in the art have read the above content, various modifications and alternatives to the present invention will be obvious. Therefore, the protection scope of the present invention should be defined by the appended claims.

Claims

An automatic test method for auditory brainstem response based on adaptive average method, characterized in that:

For ABR records collected in batches under each test sound intensity, based on the calculation of the adaptive average method, the average number of records is gradually increased through iterations until the conditions for ABR signal detection are met;

The conditions for detecting the ABR signal include: after grouping the ABR records corresponding to the current sound intensity at the current iteration, calculating the average curve of each group according to the current average times, and obtaining the value corresponding to the maximum value of the cross-correlation function between the groups Time lag, according to whether the deviation of the time lag is within the specified range, it is judged whether there is an ABR signal with time-locking characteristics.
The method for automatically testing auditory brainstem response according to claim 1, wherein:

The operation of the adaptive average method includes the following processes:

S1', the ABR record is a time curve whose input type is a repeated single record, which is randomly divided into two groups and the average curve is calculated respectively;

S2'. Calculate the cross-correlation function after the average of the two groups;

S3', obtaining the time lag corresponding to the maximum value of the cross-correlation function;

S4', compare the absolute value of the time lag with the value of k, and judge whether the time lag is within k data points:

If the absolute value of the time lag is less than the value of k, it means that the time lag deviation is within k data points, which means that the stable time lock signal is detected, and the acquisition of data is stopped, which corresponds to the detection of the ABR signal;

If the absolute value of the time lag is not less than the value of k, it means that the time lag deviation is not within k data points, indicating that the stable time lock signal is not detected, then step S5' is further executed:

S5', judge whether the maximum number of iterations under the current sound intensity is reached:

If the maximum number of iterations is reached, the acquisition of the collected data is stopped, which corresponds to the situation where the ABR signal is not detected; if the maximum number of iterations is not reached, the newly collected ABR records are further added to the currently collected ABR records, and the current sound intensity Repeat the operation process of S1'～S5';

Among them, k is a preset fixed value, or a value obtained by calculating all data points according to a preset ratio.
The method for automatically testing auditory brainstem response according to claim 1, wherein:

The operation of the adaptive average method includes the following processes:

The ABR record is a time curve whose input type is a repeated single record, which is judged Q times in parallel, and steps S1 to S4 are executed each time;

S1. The currently collected ABR records are randomly divided into two groups and the average curves are calculated respectively;

S2. Calculate the cross-correlation function after the average of the two groups;

S3. Obtain the time lag corresponding to the maximum value of the cross-correlation function;

S4. Compare the absolute value of the time lag with the k value to determine whether the time lag is within k data points:

If the absolute value of the time lag is less than the value of k, it means that the time lag deviation is within k data points, indicating that a stable time-lock signal is detected; if the absolute value of the time lag is not less than the value of k, it means that the time lag deviation is not in k data Within the point, it means that no stable lock signal is detected;

S5. Judge whether the stable lock signal is detected every time in Q parallel judgments:

If the stable lock signal is detected every time in Q times, stop acquiring the collected data, which corresponds to the situation where the ABR signal is detected;

If the stable time lock signal is not detected every time in Q times, step S6 is further executed:

S6. Determine whether the maximum number of iterations under the current sound intensity is reached:

If the maximum number of iterations is reached, stop acquiring the collected data, which corresponds to the situation that no ABR signal is detected;

If the maximum number of iterations is not reached, the newly collected ABR record is further added to the currently collected ABR record, and the calculation process from S1 to S6 is repeated under the current sound intensity;

Among them, k is a preset fixed value, or a value obtained by calculating all data points according to a preset ratio.
The method for automatically testing auditory brainstem response according to claim 1, wherein:

The operation of the adaptive average method includes the following processes:

S1 ", the ABR is that the input type is the average recording time of multiple curves; mean curve avgA this newly added plus iteration, the mean curve avgA old binding the previous iteration, the weight is calculated in accordance with the current average weights obtained new curve avgA

Among them, avgA new , avgA old , avgA plus respectively have Q groups; M is the current iteration number;

S2", new to the current average curve avgA of group Q, calculate the cross-correlation function between each two groups;

S3", obtain the Q group time lag corresponding to the maximum value of the cross-correlation function;

S4". In the Q parallel judgment, the absolute value of each group of time lag is compared with the value of k to determine whether the time lag is within k data points: if the absolute value of the time lag is less than the value of k, it means the time lag If the deviation is within k data points, it means that the stable time-lock signal is detected; if the absolute value of the time lag is not less than the k value, it means that the time-lag deviation is not within k data points, which means that the stable time-lock signal is not detected;

S5", judge whether the stable lock signal is detected every time in Q parallel judgments:

If the stable lock signal is detected every time in Q times, stop acquiring the collected data, which corresponds to the situation where the ABR signal is detected;

If the stable lock signal is not detected every time in Q times, step S6" is further executed:

S6", judge whether the maximum number of iterations under the current sound intensity is reached:

If the maximum number of iterations is reached, the acquisition of collected data is stopped, which corresponds to the situation where the ABR signal is not detected; if the maximum number of iterations is not reached, the newly collected ABR test average data is further added to the currently collected ABR test average data. Repeat the calculation process of S1"～S6" under the current sound intensity;

Among them, k is a preset fixed value, or a value obtained by calculating all data points according to a preset ratio.
The automatic test method for auditory brainstem response according to claim 2 or 3 or 4, characterized in that judging whether the maximum number of iterations under the sound intensity is reached includes judging whether M = M max ; where M max is The preset maximum number of iterations; M is the current number of iterations, and M is incremented by one in each step S1: if M< Mmax , the iteration continues.
The method for automatically testing auditory brainstem responses according to any one of claims 1 to 5, wherein the average number of times used when calculating the average curve of the ABR record at each sound intensity is the current iteration number and N The product of the values; the N value corresponds to the number of groups of ABR records newly added in each iteration.
The method for automatically testing auditory brainstem responses according to any one of claims 1 to 6, wherein the hearing threshold is obtained by detecting the minimum sound intensity required for the ABR signal; or, for each sound intensity The number of iterations, the sound intensity corresponding to the hearing threshold is obtained through function fitting and interpolation.
The method for automatically testing auditory brainstem responses according to claim 7, wherein the number of iterations used at each sound intensity is normalized, and the sound intensity corresponding to the hearing threshold is obtained through Sigmoid function fitting and interpolation. .
The method for automatically testing auditory brainstem responses according to any one of claims 1 to 5, wherein the ABR record at each sound intensity is animal experiment data or clinical data, obtained through real-time testing or offline Stored data;

The ABR record has undergone one or more of the following preprocessing:

Signal amplification

Band pass filtering;

Adjust the time interval for collecting the ABR waveform, and select the ABR time curve of the corresponding time interval after the sound stimulation starts as the analysis object;

Eliminate the ABR time curve corresponding to the background noise;

The smooth spline fitting function removes low-frequency background noise.